Update json.config to account for beginning and end of document tokens

Currently, this model fails when you pass it a token length of 514 and set truncation=True, because two additional tokens are added (beginning and end of document). Truncating input strings at 512 will prevent this error from occuring.

config.json

@@ -24,7 +24,7 @@
     "positive": 2
   },
   "layer_norm_eps": 1e-05,
-  "max_position_embeddings": 514,
+  "max_position_embeddings": 512,
   "model_type": "roberta",
   "num_attention_heads": 12,
   "num_hidden_layers": 12,